134　Yukiko Ueda

135

Identification of Null Arguments in Japanese

of Tokyo. [A revised version in, Grant-in-aid for COE research, Report (2A) Researching and verifying an advanced theory of human language: Explanation of the human faculty for constructing and computing sentences on the basis of lexical conceptual features, ed. Kazuko Inoue, 253-296. Kanda University of International Studies, 2000.]

Jun Abe 1. Introduction* This article aims to consider the syntactic status of null arguments appearing relatively freely in such a language as Japanese; more specifically, to address the question whether they are properly characterized as pro(nominal), just like null subjects exhibited in standard pro-drop languages such as Italian and Spanish. The intuition behind this issue is that the pro-drop phenomena exhibited by such languages as Chinese, Japanese and Korean, which lack overt agreement, might manifest themselves in a way quite different from those exhibited by the above-mentioned Romance languages, which are closely related to “rich agreement” in a rather obvious sense. This intuition has in fact been fleshed out in various ways in the literature. Thus, Huang (1982, 1984) argues that in a topic-prominent language such as Chinese, null topics play a crucial role in licensing null arguments and reaches the interesting conclusion that null objects are always variables bound by null topics, contrary to null subjects, which can be ordinary pronouns lacking phonetic content. Huang (1991) also claims that some instances of apparent object pro-drop phenomena in Chinese are best characterized as involving VP Ellipsis, and following this line of analysis, Otani and Whitman (1991) extend it to Japanese. More recently, Oku (1998) and Kim (1999) modify Otani and Whiman’s approach in such a way that simply NP Ellipsis is involved in the relevant phenomena. The main claim of this article is that there should be at least three ways of identification for null arguments in Japanese, each case of which is illustrated below: (i)

Discourse-based anaphora:

(1) a. [e] Kita. came ‘[e] came.’ *

I would like to express my sincere gratitude to Hiroto Hoshi, the organizer of the International Symposium on Language, Mind, and Brain, held at Akita University on October, 2006, to invite me to give a talk at this symposium. I would also like to thank the participants of this symposium, especially Ruth Kempson and Daiko Takahashi for their invaluable comments. Of course, the usual disclaimer applies.

136　Jun Abe

b. John-wa totemo [e] aisiteiru. -Top very much love ‘John loves [e] very much.’ itta.” (2) “Dare-ga Maryi -ni atta no?” “John -wa anata-ga [e]i atta to who-Nom -Dat saw Q -Top you -Nom saw Comp said ‘“Who saw Maryi? “John said you saw [e]i.”’ (ii) Intra-sentential anaphora: kuru to itta. (3) a. Johni -wa [e]i kyoo-no gogo -Top today-Gen afternoon come Comp said ‘Johni said [e]i would come this afternoon.’ motteiru ka inai ka-wa mondai b. Johni -wa [e]i okane-o -Top money-Acc have Q not Q-Top problem de-wa nai to itta. be-Top not Comp said ‘Johni said whether [e]i has money or not is not a problem.’ (iii) Anaphora involving sloppy identity: (4) John-ga zibuni-o hihansita ra, Mary -mo [e]i hihansita. -Nom self-Acc criticized and -also criticized ‘John criticized himselfi and Mary criticized [e]i.’ The following discussions proceed as follows: In Section 2, I demonstrate, following Abe (2006), that cases of anaphora involving sloppy identity, exemplified in (4), are best captured under the NP Ellipsis analysis, advocated by Oku (1998) and Kim (1999), rather than Otani and Whitman’s (1991) VP Ellipsis analysis. Then, in Section 3, I argue that the NP Ellipsis analysis cannot capture the whole phenomenon of null arguments in Japanese, demonstrating the need of other devices for identification: the null topic option for cases of discourse-based anaphora, exemplified in (1) and (2), and the option of bound pronouns for cases of intra-sentential anaphora, exemplified in (3). Finally, in Section 4, I consider some theoretical implications of the conclusions that I have drawn concerning the identification of null arguments in Japanese. The most significant conclusion that is reached in this investigation will be that Japanese null arguments do not function as referring, a property typical of pro found in standard pro-drop languages such as Italian and Spanish. Based upon this observation, I suggest that Japanese null arguments may have the size of NP, lacking the D projection, and that this will be the source of the inability of referring as well as the availability of the ellipsis strategy. This accords with

Identification of Null Arguments in Japanese　137

Fukui’s (1986) claim that the absence of functional categories (or modestly their defectiveness) plays a crucial role in the parametrization between those languages which lack agreement and those which do not.

2. Support for the NP Ellipsis Analysis Based upon Huang’s (1991) VP Ellipsis analysis to those instances of Chinese which appear to involve sloppy identity in supplying the content of null arguments, Otani and Whitman (1991) claim that such a Japanese instance as (4) above may also involve VP Ellipsis. They provide a similar example such as the following: (5) John -wa [zibun -no tegami] -o suteta. Mary -mo [e] suteta. -Top self -Gen letter -Acc discarded -also discarded ‘John threw out his letter. Mary did too.’ Such a sentence is peculiar in that the null object can be interpreted not only as ‘John’s letter’ but also as ‘Mary’s letter’. If this null object were taken as a simple pronominal, then it would be expected that it only referred to ‘John’s letter’. To account for the availability of these two readings, Otani and Whitman follow Huang (1991) in claiming that the second sentence of (5) may involve V raising to Infl, as shown in the following structure, so that the whole VP containing the null object and the trace of V can be taken as an elliptic site: (6) [IP NPsubj [I’ [VP [NP e] tverb] V-Infl]] Given that the verb of the first conjunct is also raised into the above Infl, then the VP of the first conjunct corresponds to zibun-no tegami ‘self’s letter’ and the trace of the verb. And after this VP is copied onto that of the second conjunct, we will successfully get the appropriate content for the second conjunct. Then, the availability of the two readings noted above can be accommodated by the standard analysis of VP Ellipsis constructions such as the following, which also has those two readings: (7) John threw out his letter, and Mary did too. If we follow Sag (1976) and Williams (1977), then the strict reading is derived when the pronoun his is taken as referential, coreferring to ‘John’, and the sloppy reading is derived when it is taken as a variable bound by a lambda operator. We can derive both readings available to (5) if we analyze zibun-no ‘self-Gen’ in the same way as his in (7). The strongest motivation for this VP Ellipsis analysis, according to Otani

138　Jun Abe

and Whitman, is that it can properly capture the availability of a sloppy reading in such a construction as exemplified in (4) and (5). The most interesting prediction of this approach is, then, that the availability of the sloppy reading should correlate with that of the VP Ellipsis analysis in question. Otani and Whitman claim that this is in fact the case. Thus, the following example contrasts with (4) and (5) in that it does not allow a sloppy reading: (8)

Zibun-no hatake-no ninzin-ga McGregor ozisan-no daikoobutu self-Gen garden-Gen carrot-Nom Mr. -Gen big-favorite desita. Peter -mo [e] daisuki desita. was -also very-fond-of was ‘Carrots from self’s garden were Mr. McGregor’s big favorite. Peter was also very fond of [e].’

The second sentence of this example can only mean that Peter was also very fond of carrots from McGregor’s garden. This will follow straightforwardly from Otani and Whitman’s analysis, since there is no way of analyzing the second sentence of this example as involving VP Ellipsis, lacking the antecedent VP to be copied. Given this argument, the validity of Otani and Whitman’s claim that such examples as (4) and (5) may involve VP Ellipsis should be examined on the basis of whether the correlation between the availability of the sloppy reading and that of the VP Ellipsis analysis is real. Oku (1998) and Abe (2006) demonstrate that it is not. Though it is true that such an example as (8) lacks a sloppy reading, there is another account for this fact that is no less plausible, as Otani and Whitman themselves mention briefly; that is, the availability of the sloppy reading requires “some type of parallelism constraint.” It would not be unreasonable to claim that unlike (4) and (5), (8) does not satisfy the parallelism constraint in question. Otani and Whitman would claim that their VP Ellipsis analysis ensures how such a parallelism constraint is satisfied in order to get a sloppy reading. The question, then, comes down to whether the parallelism constraint in question is exactly what the VP Ellipsis analysis demands. A good case to examine this question is one in which the coordinated sentences appear to meet such a parallelism requirement and yet it is null subjects that are involved in producing sloppy readings. In Abe (2006), I provide the following examples:1 (9) a. Zibun-no hatake-no ninzin-ga McGregor ozisan-no daikoobutu self -Gen garden-Gen carrot -Nom Mr. -Gen big-favorite 1



Oku (1998) also provides similar examples to make the same point.

Identification of Null Arguments in Japanese　139



desita. [e] Peter -no daikoobutu de -mo arimasita. was -Gen big-favorite -also was ‘Carrots from self’s garden were Mr. McGregor’s big favorite, and [e] was Peter’s big favorite, too.’ b. John -wa zibun-no tuma-ga kiree da to omotteiru. -Top self -Gen wife-Nom beautiful be Comp think Bill-mo [e] kiree da to omotteiru. -also beautiful be Comp think ‘John thinks self’s wife is beautiful. Bill also thinks [e] is beautiful.’ I then claim that in both these examples, the second sentences allow sloppy readings, and that according to Otani and Whitman’s analysis, these examples should not allow sloppy readings, since they do not involve configurations that can be analyzed as VP Ellipsis constructions. Thus, to the extent that the facts exemplified in (9) are real, they undermine the validity of Otani and Whitman’s analysis considerably. Based upon such a fact as mentioned above as well as others, Oku (1998) and Kim (1999) argue convincingly against Otani and Whitman’s (1991) VP Ellipsis analysis for null object constructions, and propose instead that such null object constructions as considered by Otani and Whitman simply constitute part of the more general phenomena of null arguments, and that they all involve NP Ellipsis. Note that this proposal gains strong support from the facts mentioned above, since they demonstrate that not only null objects but also null subjects can have sloppy readings and also that no restriction that would be imposed under the VP Ellipsis analysis seems to be at work in the availability of such a reading, the only obvious restriction being some sort of parallelism constraint. Notice that the mechanism that is involved in deriving strict and sloppy readings under the VP Ellipsis analysis can be carried over to the NP Ellipsis analysis with no significant change. Crucially, Oku (1998) and Kim (1999) share with Otani and Whitman the idea that null arguments of these languages are not simply empty pronouns. This point is most clearly illustrated by (4) or a similar example such as the following: (10) Hotondo subete-no hito-ga zibun-o hihansita ga, almost every -Gen person-Nom self -Acc criticized but John -dake-wa [e] hihansinakatta. only-Top didn’t-criticize ‘Almost everyone criticized self, but only John didn’t criticize [e].’

140　Jun Abe

The most natural interpretation of the second sentence of this example is the sloppy reading according to which only John did not criticize himself. If the null object of the second sentence were simply an empty pronoun, then it would violate Condition B of the binding theory. Let us compare this example with the following: (11) John-wa [e] hihansita. -Top criticized i) * John criticized himself. ii) John criticized someone else. Interestingly, this sentence does not have the reading in which John criticized himself when uttered out of the blue. The contrast between (10) and (11) with respect to the availability of the self reading suggests that only when zibun ‘self’ is supplied in a context such as (10) is the self-reading available. Kim (1999) demonstrates that the NP Ellipsis analysis gives a straightforward account to this contrast. In (11), the only antecedent for the null object is ‘John’ and hence after it is copied onto the null object, it induces a Condition C violation.2 On the other hand, in (10), zibun can be copied onto the null object of the second sentence, thereby giving rise to the self reading. Further evidence for the NP Ellipsis analysis can be given with the phenomenon of split antecedent in elliptic constructions, discussed by Lobeck (1995). Consider the following examples, taken from Lobeck (1995, p. 29-30): (12) a. Wendy is eager to sail around the world and Bruce is eager to climb Kilimanjaro, but neither of them can [e] because money is too tight. (Webber 1978) b. John wants to climb Kilimanjaro and Bill would like to scale Everest, but neither of them knows why [e]. (12a) illustrates an instance of VP Ellipsis and (12b) an instance of Sluicing. The interesting property of these examples is that the elliptic sites involve split antecedent. Thus, the second conjunct of (12a) is interpreted as follows: Wendy cannot sail around the world and Bruce cannot climb Kilimanjaro because 2

However, this account raises the question why the full sentence which results from copying John onto the null object in (11) is almost perfect in its acceptability, as shown below: (i) John -wa John -o hihansita. -Top -Acc criticized ‘John criticized John.’ See Section 3.2 for relevant discussion.

Identification of Null Arguments in Japanese　141

money is too tight. The same way of interpretation also obtains in (12b). Now, this phenomenon of split antecedent is also observed in null arguments in Japanese, as illustrated below: (13) Mary -wa John -to kekkon-si, Susan-wa Bill-to rikon-sita ga, -Top -with married -Top -with divorced but a. dotiramo [e] kuyandeiru. both regret b. dotiramo [e] matigai datta to omotteiru. both mistake was Comp think ‘Mary married John and Susan divorced Bill, and (a) both regret [e]. (b) both think that [e] was a mistake.’ Both continuations (a) and (b) allow the interpretations of split antecedent. Thus, (13a) can be most naturally interpreted as follows: Mary regrets that she married John and Susan regrets that she divorced Bill. (13b) can also be interpreted in a similar way. To the extent that such a phenomenon of split antecedent is a property peculiar to the ellipsis construction in general, the availability of such a reading for null arguments in Japanese, as illustrated in (13), lends support to an analysis of such null arguments as involving ellipsis. Furthermore, the fact that such a reading is equally available to both null objects and subjects, as shown in (13a) and (13b), respectively, lends strong support to the NP Ellipsis analysis rather then Otani and Whitman’s (1991) VP Ellipsis analysis. Further support for the NP Ellipsis analysis can be obtained from Hoji’s (1998) interesting observations with respect to what portion of an NP functions as the antecedent of a null argument in Japanese. Consider one of his examples, given below: (14) A: John -ga zibun-no kuruma-o aratta. -Nom self -Gen car-Acc washed ‘John washed his car.’ B: Bill -mo [e] aratta. -also washed ‘Bill also washed [e].’ Hoji observes that in this dialogue, B’s utterance can be interpreted in many ways, as given below: (15) Bill also washed a car/John’s car/Bill’s car. Based upon such facts, Hoji suggests that what is copied onto a null argument

142　Jun Abe

Identification of Null Arguments in Japanese　143

may be simply the N head of its antecedent and that “a nominal projection whose sole content is its head N can be interpreted in various ways as just indicated,” (p. 142) for example in (15). It seems, however, that these facts simply show the possibility of copying only the N head of the antecedent of a null argument, but not the obligatoriness of such a way of copying, since when only the head of zibun-no kuruma ‘self’s car’ in (14A), namely kuruma ‘car’, is supplied to the null object in (14B), the resulting sentence does not have all the readings given in (15), as shown below: (16) A: John-ga zibun-no kuruma-o aratta. -Nom self -Gen car-Acc washed ‘John washed his car.’ B: Bill -mo kuruma-o aratta. -also car-Acc washed ‘Bill also washed a car.’ It seems to me that (16B) allows only the indefinite reading of a car, which clearly contrasts with such a case as (14B). To the extent that the judgment is clear enough, it is natural to conclude that the indefinite reading of a car in (14B) is derived by copying the N head of the antecedent of the null object, but the other two readings should be derived in the standard way, that is, by copying the whole NP of the antecedent of the null object. This state of affairs will be most naturally accommodated by Fukui and Spea’s (1986) theory of phrase structure, according to which lexical categories such as N, A and V have recursive projections of X’ with no closure of XP whereas functional categories such as D, T and C can be projected into XP when a specifier is supplied as an agreeing element with its head. Given the further assumption made by Fukui (1986) that in such a language as Japanese which lacks agreement, no functional category exists (or more modestly, functional categories are defective), it is reasonable to claim that null arguments in Japanese can take any N’ projections as the targets of copying. Suppose that zibun-no kuruma ‘self’s car’ has the following structure: (17)

N’

N’ N’ | | self’s N | car

Then, either the whole N’ or the smaller N’ including only ‘car’ can be copied onto the null object in (14B), so that it can give rise to not only the sloppy and strict readings but also the indefinite reading.3 Given that any portion of N’ projection can serve as an antecedent of a null argument in Japanese, this will lend support to the NP Ellipsis analysis, since this fact is easily accommodated under this analysis while it is hard to imagine how it is accommodated under Otani and Whitman’s (1991) VP Ellipsis analysis.

3. Problems of the NP Ellipsis Analysis I have demonstrated that the device of NP Ellipsis plays a crucial role in identifying null arguments in Japanese. Nonetheless, it will be shown in this section that this device cannot be the only one to capture the whole phenomenon of null arguments in this language and that there are at least two other devices necessary for the identification in question: variables bound by null topics, as invented by Huang (1982, 1984), and bound pronouns in need of antecedents.

3.1 Cases of Discourse-based Anaphora: The Need of Null Topics It is rather obvious that the device of NP Ellipsis cannot be the only one for identifying null arguments in Japanese, given those cases in which there are no linguistic antecedents for null arguments and their reference is determined in terms of prominency in a given discourse, as shown below: (18) a. In the situation in which the speaker and hearer are expecting somebody to come soon: Hora [e] kita. look came ‘Look! [e] came.’ b. In the situation in which the speaker is pointing to John playing with his pet: John -wa totemo [e] kawaigatteiru. -Top very much cherish ‘John is cherishing [e].’ Hankamer and Sag (1976) show that VP Ellipsis and Sluicing constructions need linguistic antecedents, as exemplified below: 3

Note that this account is not compatible with the bare phrase structure theory proposed by Chomsky (1994). To make it compatible with this theory, we will need to claim that the absence of D makes it possible for any projection of N to serve as an antecedent of a null argument. See Section 4 for relevant discussion.

144　Jun Abe

(19) [Sag produces a cleaver and prepares to hack off his left hand] Hakamer: #Don’t be alarmed, ladies and gentlemen, we’ve rehearsed this act several times, and he never actually does. (20) [Hankemer produces a gun, points it offstage and fires, whereupon a scream is heard] Sag: #Jesus, I wonder who. Given these data, it is natural to reason that the device of NP Ellipsis will not do for identifying the null arguments in (18).4 There are at least two candidates for identifying such arguments that immediately come to mind. One is the standard pro analysis, since such a use of null arguments as exemplified in (18) is typical in standard pro-drop languages such as Italian and Spanish, though in the latter languages, the position that is licensed as pro is usually confined to subject due to the restriction of rich agreement. The other is the null topic analysis proposed by Huang (1982, 1984), who claims that in topic-prominent languages such as Chinese and Japanese in the sense of Li and Thompson (1976), null topics are freely available and hence null arguments can be identified as variables bound by them. Thus, according to this analysis, each sentence of (18) roughly has the following representation: (21) a. [Øtopic(i) [[e](i) came]] b. [Ø topic(i) [John is cherishing [e](i) very much]] In these cases, the values of the null topics are supplied by prominent entities in a given discourse. Huang (1982) points out two reasons for not taking the standard pro analysis. One comes from his intuition that “the so-called pro-drop phenomenon happens only in languages with sufficiently rich inflection,” (p. 356~7) and hence the pro-drop phenomenon that is observed in such languages as Chinese and Japanese which do not exhibit such inflection at all should be attributed at least partly to a different mechanism. The other reason is based upon empirical facts; consider the following Japanese examples, which are adapted from Huang’s (1982) Chinese examples: itta.” (22) “Dare-ga Maryi-ni atta no?” “John-wa anata-ga [e]i atta to who-Nom -Dat saw Q -Top you-Nom saw Comp said 4

However, see Chao (1987) and Lobeck (1995) for claiming that VP Ellipsis and Sluicing constructions do not require linguistic antecedents. Given this discrepancy, the existence of such data as in (18) alone might not suffice to establish the need of an identification device other than NP Ellipsis.

Identification of Null Arguments in Japanese　145

‘ “Who saw Maryi? “John said you saw [e]i.”’ itta. (23) * Johni -wa anata-ga [e]i atta to -Top you-Nom saw Comp said ‘Johni said you saw [e]i.’ (22) is a case where a null object picks up its antecedent from a linguistic context, whereas (23) is a case where a null object pick up its antecedent intrasententially. Interestingly, as first noted by Kuroda (1965) with such a Japanese example as (23), the embedded null object cannot take the matrix subject as its antecedent. If we took the standard pro analysis, this would be unexpected; nothing would seem to prevent both null arguments in (22) and (23) from having the anaphoric relations indicated. Under the null topic analysis, on the other hand, there is a way to capture the contrast between (22) and (23), Huang claims. Given this analysis, it is natural to posit a null topic at the top of the second sentence of (22), as shown below: (24) [Ø topic(i) [John said you saw [e](i)]] Here, the null topic picks up Mary as the most prominent entity in the discourse. Huang then claims that universal principles dictate that a genuine pro be licensed only in subject position, hence explaining the ungrammaticality of (23) as well as the grammaticality of the following sentence, in which a null argument appears in the embedded subject: itta. (25) Johni -wa [e]i anata-ni atta to -Top you -Dat saw Comp said ‘Johni said [e]i saw you.’ Hence, to the extent that subject-object asymmetry holds in the availability of null arguments, as shown in (23) and (25), we have a good reason to assume that the standard pro analysis does not hold in such a language and that a special device such as a null topic plays a crucial role in identifying a null object that picks up its value outside the sentence, as shown in (22). However, whether the subject-object asymmetry in question is a real phenomenon has been controversial in the Japanese literature; while Hasegawa (1984/5) follows Huang (1982, 1984) in taking it as real, Hoji (1985) considers that such a sentence as (23) becomes acceptable once it is appropriately manipulated, indicating that the apparent unacceptability of (23) should not be ascribed to grammatical constraints. I will not commit myself to this issue in this article, but provide independent evidence for the need of the null topic analysis in identifying null arguments that

146　Jun Abe

pick up their values from a discourse. Discussing what he calls quantificational null objects in Japanese, Takahashi (2008) provides the following examples: (26) a. b.

Zyosi-no dareka-ga taitei-no sensei-o sonkei-siteiru. girl -Gen someone-Nom most-Gen teacher-Acc respect ‘Some girl respects most teachers.’ Dansi-no dareka-mo [e] sonkei-siteiru. boy -Gen someone-also respect ‘Some boy also respects [e].’

As is well-known, the scope interaction in Japanese respects the rigidity condition, according to which a structurally higher QP necessarily takes scope over a lower QP. Thus, in (26a), zyosi-no dareka ‘some girl’ must take scope over taitei-no sensei ‘most teachers’. As for (26b), Takahashi (2008) observes that it can be ambiguous depending upon what is supplied to the content of the null object. One reading will be derived from literal copying of taitei-no sensei onto the null object, following the NP Ellipsis analysis. The other amounts to a strict reading, derived by means of copying the referential value of taitei-no sensei onto the null object, hence referring to the set denoted by the domain of ‘most teachers’. Notice that this latter reading can also be derived by means of either a null topic or a standard pro, in which case they function like E-type pronouns in the sense of Evans (1980). Takahashi further observes that in (26b) the subject must take scope over the null object on either reading, as is expected from the rigidity condition working in a way parallel to the first conjunct in (26a). With this much in background, Takahashi provides the following data, which contrast in an interesting way with those given in (26): (27) a. b.

Taitei-no sensei-o zyosi-no dareka-ga sonkei-siteiru. most-Gen teacher-Acc girl -Gen someone-Nom respect ‘Some girl respects most teachers.’ Dansi-no dareka-mo [e] sonkei-siteiru. boy -Gen someone-also respect ‘Some boy also respects [e].’

These data minimally differ from those given in (26) in that (27a) involves scrambling of the object to the top of the sentence. As has been well-known since Kuroda (1971), when the object QP is scrambled over the subject QP, it makes the sentence scopally ambiguous. Thus, (27a), unlike (26a), can have the reading in which the object taitei-no sensei ‘most teachers’ takes scope over the subject zyosi-no dareka ‘some girl’. As for (27b), Takahashi makes

Identification of Null Arguments in Japanese　147

the interesting observation according to which this sentence is also scopally ambiguous; that is, it allows the reading in which the null object takes scope over the subject QP. This is unexpected if the null object remains in its original position. Hence, neither the NP Ellipsis analysis nor the standard pro analysis will account for the scope ambiguity in (27b). Takahashi suggests the possibility of scrambling the null object to the top of the second conjunct in a way parallel to the first conjunct, but there is a further interesting fact about (27b), which seems to me to be real, contra Takahashi’s judgment; that is, when the null object takes scope over the subject, it necessarily functions like an E-type pronoun. This will be unexpected, even if we admit the option of scrambling a null object; under such an analysis, the reading derived from literal copying of taitei-no sensei should also be available. Under the null topic analysis, on the other hand, the facts fall into place: The reading derived from literal copying of taitei-no sensei has the LF representation of S-O-V order, thereby giving rise to only the interpretation in which the subject takes scope over the object. The E-type reading, on the other hand, can be derived from a structure in which a null topic is involved; hence (27b) can have the following structure: (28) [Ø topic(i) [dansi-no dareka-mo [e](i) sonkei-siteiru]] In this structure, the null topic is structurally higher than the subject QP dansino dareka ‘some boy’, hence allowing the former taking scope over the latter.5 That a topic can in fact take scope over the subject just below it is shown by the following sentence, in which the null topic in (28) is replaced by the overt one: (29) Taitei-no sensei-wa dansi-no dareka-mo sonkei-siteiru. most-Gen teacher-Top boy -Gen someone-also respect ‘As for most teachers, some boy also respects them.’ Thus, Takahashi’s (2008) data provide strong support to the need of the null topic analysis.6 Given the options of NP Ellipsis and null topic, there is no need 5

Note that a parallelism constraint is at work here between the first and second conjuncts in scope ambiguity. Thus, the reading in which the null topic takes scope over the subject in (28) is available only when (27a) has a parallel reading, that is, the one in which the scrambled object takes scope over the subject. Note also that this constraint accounts for why (26b) cannot have the reading in which the null topic takes scope over the subject in a way similar to (27b). 6 Takahashi (2008) provides another piece of evidence for the possibility of scrambling null objects which is concerned with the remedy of weak crossover violation: (i) a. Taitei-no sensei-o [e yasasiku osieta gakusei]-ga sonkei-siteiru. most -Gen teacher-Acc gently taught student-Nom respect ‘Most teachers, the students [e] taught gently respect.’

148　Jun Abe

to add a genuine pro to the devices of identifying null arguments in Japanese, as far as the data considered so far are concerned, though it is also true that no data observed so far prohibit pro from being counted among the devices.

3.2 Cases of Intra-sentential Anaphora: The Need of Bound Pronouns We have not considered yet those cases in which null arguments refer intrasententially, such as given in (25), repeated here: itta. (30) Johni -wa [e]i anata-ni atta to -Top you-Dat saw Comp said ‘Johni said [e]i saw you.’ Recall that Huang (1982, 1984) claims that such a null argument is an instance of genuine pro. In the present context, we need first to examine whether such a null argument can be identified as an instance of NP Ellipsis or as a variable bound by a null topic. The latter option is unlikely since the essential property of null topics is to refer to an entity in a given discourse. As for the option of NP Ellipsis, no obvious reason has been provided to prevent it from operating in such a case as (30); in fact, Kim (1999) seems to intend to accommodate such a case as an instance of NP Ellipsis, though she does not deal with it in an explicit way. In this section, I demonstrate that those null arguments involving intrasentential anaphora cannot be identified by means of NP Ellipsis, and claim that they should be identified as bound pronouns. Recall first that under the NP Ellipsis analysis, literal copying of a quantifier onto a null argument is possible inter-sententially, as illustrated in (26), repeated below: (31) a. Zyosi-no dareka-ga taitei-no sensei-o sonkei-siteiru. girl -Gen someone-Nom most-Gen teacher-Acc respect ‘Some girl respects most teachers.’

b. [e kibisiku sikatta gakusei]-mo e sonkei-siteiru. strictly scolded student-also respect ‘The students [e] scolded strictly respect [e], too.’ I agree with Takahashi on the point that (ib) is grammatical with the reading in which the embedded null object is bound by taitei-no sensei ‘most teachers’, and hence it looks like the matrix null object remedies weak crossover violation. On the other hand, it seems to me that here again, (ib) is grammatical with the bound reading in question only if the null object acts as an E-type expression, hence referring to the same set of teachers that taitei-no sensei ‘most teachers’ in (ia) refers to. To the extent that this is the right observation, it also lends support to the null topic analysis developed in the text.

Identification of Null Arguments in Japanese　149

b. Dansi-no dareka-mo [e] sonkei-siteiru. boy -Gen someone-also respect ‘Some boy also respects [e].’ (31b) has the reading in which some boy also respects most teachers, and this reading is derived by copying taitei-no sensei-o ‘most teachers’ in (31a) onto the null object in (31b) under the NP Ellipsis analysis. It is then predicted that if this analysis also accommodates such a case as (30), literal copying of a quantifier onto a null argument should also be possible for such a case. This prediction is not borne out, however; consider the following example: (32) Taitei-no sensei-ga [e] kyoo-no gogo kuru to itta. most -Gen teacher-Nom today-Gen afternoon come Comp said ‘Most teachers said [e] would come this afternoon.’ This sentence does not have the reading equivalent to that of the following sentence, in which the two occurrences of taitei-no sensei ‘most teachers’ do not have to share the set they denote: (33)

Taitei-no sensei-ga taitei-no sensei-ga kyoo-no gogo most-Gen teacher-Nom most-Gen teacher-Nom today-Gen afternoon kuru to itta. come Comp said ‘Most teachers said most teachers would come this afternoon.’

This reading would be derived if literal copying of taitei-no sensei-ga ‘most teachers’ onto the embedded null subject was possible. That it is unavailable to (32) clearly indicates that the option of NP Ellipsis is not at work here. It is incorrect, however, to draw from the above data the conclusion that the NP Ellipsis strategy is inoperative at all for intra-sentential anaphora. Compare the following data with (32)-(33): (34) John -wa taitei-no sensei-o kiratteiru node, [e] hinansita. -Top most-Gen teacher-Acc hate because criticized ‘Because John hates most teachers, [he] criticized [e].’ (35) John -wa taitei-no sensei-o kiratteiru node, taitei-no -Top most-Gen teacher-Acc hate because most-Gen sensei-o hinansita. teacher-Acc criticized ‘Because John hates most teachers, [he] criticized most teachers.’ In (34), the antecedent of the null object appears within an adjunct clause and

150　Jun Abe

unlike that in (32), this null object can be interpreted as ‘most teachers’, so that (34) can have the reading equivalent to that of (35).7 The same pattern obtains in those cases in which the antecedents of null arguments are embedded within relative clauses; consider the following examples: (36) (37)

Taitei-no sensei-o kiratteiru gakusei-ga [e] hinansita. most-Gen teacher-Acc hate student-Nom criticized ‘The students who hate most teachers criticized [e].’ Taitei-no sensei-o kiratteiru gakusei-ga taitei-no sensei-o most-Gen teacher-Acc hate student-Nom most-Gen teacher-Acc hinansita. criticized ‘The students who hate most teachers criticized most teachers.’

Here again, the null object in (36) can be identified as ‘most teachers’, so that this sentence can be interpreted in exactly the same way as (37). How can we capture the difference observed above with respect to the availability of literal copying of quantifiers onto null arguments under the NP Ellipsis analysis? A key notion that immediately comes to mind is the c-command relation holding between null arguments and their antecedents. With this notion, we can give the following descriptive generalization: (38) NP Ellipsis does not apply to a null argument when it is c-commanded by its antecedent. This generalization captures the above facts correctly: (32) cannot have the reading which would be derived if the quantifier antecedent of the null embedded subject is literally copied onto it, since in this case, the antecedent c-commands the null argument, thereby making the NP Ellipsis strategy

Identification of Null Arguments in Japanese　151

unavailable. (34) and (36) do have the readings in question, since in these cases, the c-command relation does not hold between the null arguments and their antecedents, hence not preventing the NP Ellipsis strategy from applying to them.8 More examples of a different sort can be provided to demonstrate that (38) holds true. Consider first the following example: (39) John -wa zibun-no musume-o kiratteiru ga, Bill -wa [e] suiteiru. -Top self-Gen daughter-Acc hate but -Top like ‘John hates his own daughter but Bill likes [e].’ This sentence has the reading in which the second conjunct is interpreted as ‘Bill likes his own daughter’, and this reading is derived under the NP Ellipsis analysis by literal copying of zibun-no musume ‘self’s daughter’ onto the empty object of the second conjunct. Now compare (39) with the following sentence: (40) John -wa zibun-no musume-ni sensei-ga [e] ai-tagatteiru to itta. -Top self-Gen daughter-Dat teacher-Nom want-to-see Comp said ‘John told his own daughter that the teacher wanted to see [e].’ This sentence does not have the reading which would be derived by literal copying of zibun-no musume ‘self’s daughter’ onto the null argument, so that ‘self’s daughter’ may be interpreted as ‘the teacher’s own daughter’. This fact lends support to the correctness of the generalization given in (38), since in (40), the null argument is c-commanded by its antecedent, unlike that in (39). The 8

Ruth Kempson (personal communication) raises the question whether the notion of c-command is the right notion to capture the availability of the NP Ellipsis strategy, pointing out that one of the good cases to test this claim is one in which the antecedent of a null argument is embedded in the possessive position of an NP. Let us compare the following examples:

7

Hasegawa (1984/5) reaches the conclusion that when the antecedent of a null argument is within an adjunct clause, it cannot be an indefinite or non-specific NP, providing such examples as the following (I slightly modified the original examples): (i) a.(*)John -wa [ hurui kuruma -ga sukida kara] [e] katta. -Top old car -Nom like because bought ‘Because he likes old cars, John bought [e].’ b.(?*)John -wa [usagi-ga hatake-o arasita kara] [e] kiratteiru. -Top rabbit-Nom field-Acc damaged because hate ‘Because rabbits did harm to the field, John hates [e].’ Though Hasegawa judges these sentences as basically unacceptable, as indicated in the parenthesized marks, I find them as fairly normal with the interpretations in which the null objects are identified as ‘old cars’ and ‘rabbits’, respectively, which will be derived by literal copying of their antecedents under the NP Ellipsis strategy.

(i) a. Takusan-no gakusei-ga [e] sotugyou-siki-ni syusseki-suru to omotteiru. many -Gen student-Nom graduation ceremony-Dat attend Comp think ‘Many students think that [e] will attend the graduation ceremony.’ b. Takusan-no gakusei-no sidou-kyouin-ga [e] sotugyou-siki-ni many -Gen student-Gen major advisor-Nom graduation ceremony-Dat syusseki-suru to omotteiru. attend Comp think ‘Many students’ major advisors think that [e] will attend the graduation ceremony.’ (ia) does not have the reading which would be derived by literal copying of takusan-no gakuseiga ‘many students’ onto the embedded null subject, just like (32). It seems to me that (ib) also lacks the reading in question. This might indicate that the notion of c-command relevant here needs to be weakened in such a way that an element occupying the specifier position of an NP can c-command whatever the NP does.

152　Jun Abe

same pattern of facts holds with reciprocals. Compare (41) with (42). (41) John to Mary -wa otagai-no musume-o kiratteiru ga, and -Top each other-Gen daughter-Acc hate but Bill to Susan -wa [e] suiteiru. and -Top like ‘John and Mary hate each other’s daughters but Bill and Susan like [e].’ (42) John to Bill -wa otagai-no musume-ni sensei-tati-ga [e] and -Top each other-Gen daughter-Dat teacher-pl-Nom ai-tagatteiru to itta. want-to-see Comp said ‘John and Bill told each other’s daughters that the teachers wanted to see [e].’ (41) has the reading in which the second conjunct is interpreted as ‘Bill and Susan like each other’s daughters’, which is straightforwardly derived by literal copying under the NP Ellipsis analysis. (42), on the other hand, does not have the corresponding reading, which would be derived by literal copying of otagai-no musume ‘each other’s daughters’ onto the null argument, so that the reciprocal may take ‘the teachers’ as its antecedent. These facts also demonstrate the correctness of the generalization of (38). Further, compare (40) and (42) with the following examples: (43) (44)

Zibun-no musume-o kiratteiru hito-ga Bill-ni [e] aise yo to self -Gen daughter-Acc hate person-Nom -Dat love Comp tyuukosusita. advised ‘A person who hates his/her own daughter advised Bill to love [e].’ Otagai-no musume-o kiratteiru hutari-ga Mary to Susan -ni [e] each other-Gen daughter-Acc hate two-Nom and -Dat aise yo to tyuukokusita. love Comp advised ‘Two persons who hate each other’s daughters advised Mary and Susan to love [e].’

Contrary to (40) and (42), these sentences have the readings in which the null objects are interpreted as ‘Bill’s own daughter’ and ‘each other’s daughters’ taking Mary and Susan as the antecedent of ‘each other’, respectively. Again, these facts lend support to the correctness of the generalization of (38), since in these cases, the antecedents of the null objects are embedded within relative

Identification of Null Arguments in Japanese　153

clauses and hence do not c-command them, thereby susceptible to the NP Ellipsis analysis. It remains to address the question why the generalization given in (38) holds, but at this point, I do not have much to say about this question but to speculate that (38) can be extended as a universal principle holding for ellipsis in general, to such an effect as the following: (45) Ellipsis cannot hold when the antecedent c-commands the elliptic site. It is true that in all cases of ellipsis standardly known as such, including VP Ellipsis, Sluicing, N’-Deletion, Gapping, Stripping, etc., the c-command relation does not hold between the elliptic site and its antecedent, and hence, as far as I can see, (45) will not contradict the relevant empirical facts. If the generalization given in (38) is the right way to capture the facts in question, then it suggests that there is a universal constraint to the effect stated in (45) which happens to have been concealed due to lack of positive evidence before the NP Ellipsis analysis was proposed. Assuming that the generalization stated in (38) or (45) holds true, it still remains to address the question how a null argument is identified when it is c-commanded by its antecedent. One may naturally claim that this is in fact a case of genuine pro. Given that such identification is necessary only when a null argument is bound by its antecedent, one may also wonder whether such an instance of pro should be regarded as the same entity as those of pro postulated for such standard pro-drop languages as Italian and Spanish. Putting aside detailed discussions of this issue until the next section, let us consider another case of null arguments that demonstrates that they need to be identified as bound pros. In so doing, let us first review Williams’s (1977) account of the following case of VP Ellipsis: (46) a. John saw everyone before Mary did [e]. b. Yes, and Bill did [e] before Sally did [e]. (46a) is two-ways ambiguous: One reading is derived by literal copying of the VP see everyone onto the VP elliptic site, hence fitting into the situation where John saw everyone and then Mary saw everyone. The other reading is derived from the following LF representation, where everyone undergoes QR, adjoining to the matrix TP: (47) [TP everyone [TP John Past [VP see t] before Mary did [VP e]]] In this representation, the VP see t can be copied onto the null VP, hence giving

154　Jun Abe

rise to the reading according to which for each person x, John saw x before Mary saw x (notice that this does not entail that John saw everyone before Mary saw everyone). Williams then notes that contrary to (46a), (46b) has only the first reading. This reading is derived by simply copying the VP see everyone in (46a) onto both occurrences of the empty VPs. The reason why (46b) lacks the second reading is attributed to the crucial assumption made by Williams that LF copying must take place after QR has applied; while the latter is an operation that derives an LF representation, the former is an operation applying at the output of the LF component. Thus, we could derive the reading in question by copying the VP see everyone in (46a) onto the first occurrence of the empty VPs in (46b), applying QR to everyone to give rise to a representation similar to (47), and then copying the VP see t onto the second occurrence of the empty VPs, but this violates the constraint on rule ordering just mentioned. With this much in mind, let us consider the Japanese counterpart of (46) which involves null objects rather than VP Ellipsis, as shown below: (48) a. John-wa daremo-ni [Mary-ga [e] au mae-ni] atta. -Top everyone-Dat -Nom see before saw ‘John saw everyone before Mary saw [e].’ b. Sosite, Bill -mo [e] [Sally -ga [e] au mae-ni] atta. and -also -Nom see before saw ‘And Bill saw [e] before Sally saw [e].’ (48a) is two-ways ambiguous, just like (46a). The first reading is derived by literal copying of daremo-ni ‘everyone’ onto the null object. As for the second reading, if we put aside the restriction given in (38) or (45), it will be derived from an LF representation corresponding to (47) by copying the trace of daremo-ni onto the null object.9 Interestingly, (48b) is also two-ways ambiguous, contrary to (46b), and the question is why it has the second reading. If the NP Ellipsis analysis was the only option to take for this case, then it would predict the unavailability of this reading to (48b) for exactly the same reason as (46b) lacks it. On the other hand, if the option of a bound pro is admitted in this case, then the availability of the second reading follows immediately; the content of the first null object is obtained by literal copying of daremo-ni ‘everyone’ and the second null object simply serves as a pro bound by this quantifier.10 Turning 9

See Takahashi (2006) for the argument that traces can be copied under the NP Ellipsis analysis. 10 Given the ordering constraint on QR and LF copying, we need to assume that a quantifier does not have to undergo QR to serve as an antecedent of a bound pro. This might get along well with the current conception of QR that it applies only when necessary to obtain an

Identification of Null Arguments in Japanese　155

back to the availability of the reading in question to (48a), this case alone can be regarded as lending support to the existence of a bound pro, given the restriction stated in (38) or (45); since daremo-ni ‘everyone’ c-commands the null object, the NP Ellipsis analysis must be inoperative in deriving the reading in question. That the c-command relation is a crucial factor for the availability of this reading is shown by the contrast between (48a) and the following sentence, in which daremo-ni is put after the adjunct clause: (49) John -wa [Mary -ga [e] au mae-ni] daremo-ni atta. -Top -Nom see before everyone-Dat saw ‘John saw everyone before Mary saw [e].’ This sentence lacks the reading in question, unlike (48a), and this is naturally attributed to the lack of the c-command relation between daremo-ni and the null object.11 It is further demonstrated that a bound pro identified as such under the present analysis obeys Condition B of the binding theory. Recall that we have observed that (11), repeated here as (50), does not have the reading in which the null object is interpreted as ‘John’. (50) John -wa [e] hihansita. -Top criticized Recall also that Kim (1999) gives an account to this fact under the NP Ellipsis analysis: the only antecedent for the null object is ‘John’ and hence after it is copied onto the null object, it induces a Condition C violation. However, as noted in footnote 2, this account raises a problem; that is, why the full sentence which results from copying John onto the null object in (50) is almost perfect in its acceptability, as shown below: (51) John -wa John -o hihansita. -Top -Acc criticized ‘John criticized John.’ The present analysis, on the other hand, does not induce such a problem. Since the null object is c-commanded by ‘John’ in (50), the NP Ellipsis analysis is inoperative here. The only option left is to identify it as a pro bound by ‘John’, otherwise underivable reading, advocated by Fox (2000) and Reinhart (1998). 11 This in turn indicates that (48a) is structurally ambiguous depending on where the adjunct clause is adjoined, since the availability of the first reading to (48a) leads us to conclude that in the structure that derives this reading, the antecedent of the null object of au ‘see’, i.e., daremoni ‘everyone’ does not c-command it, so that the NP Ellipsis strategy is available.

156　Jun Abe

but this induces a Condition B violation; hence the unavailability of the reading in which John criticized himself. Further evidence can be provided to make the same point; consider first the following example (the sentences given in (52) and (53) are created on the model of the Italian examples given in Rizzi (1986)): (52) [[e] [ syatyoo-ga [e] hataraku koto-o kyoosei dekiru] koto-o president-Nom work fact-Acc force can fact-Acc mitomeru] no -wa yoku nai. admit -Top good not ‘It is not good that [e] admits that a boss can force [e] to work.’ Since this sentence makes a general statement, the two occurrences of null arguments indicated as [e] in (52) refer to arbitrary persons. The crucial fact in the present context is that the second null argument can take the same value as the first, so that the agent of admitting may be the same person as the recipient of forcing. Now compare (52) with the following sentence: (53)

Kono kaisya-de-wa [[e] [e] hataraku koto-o kyoosei-suru] koto-wa this company-in-Top work fact-Acc force fact-Top muzukasii. difficult ‘In this company, it is difficult that [e] forces [e] to work.’

This sentence also makes a general statement and the two occurrences of null arguments indicated as [e] in (53) refer to arbitrary persons. Contrary to (52), however, these two null arguments cannot take the same value, so that the agent of forcing must be different from the recipient of forcing. This fact immediately follows under the present analysis, according to which the null object of the verb ‘force’ is regarded as a pro bound by the null subject of the same verb, hence inducing a Condition B violation. Note in passing that we have not discussed how a null argument obtains arbitrary interpretation, to begin with; I will make relevant discussion in the next section. To summarize, we found out in this subsection that as far as intra-sentential anaphora is concerned, null arguments in Japanese behave like pro, obeying Condition B of the binding theory, when they are bound by their antecedents, whereas they behave like real empty elements that need to be supplied with their content by the NP Ellipsis strategy when they are not bound by their antecedents.

Identification of Null Arguments in Japanese　157

4. Theoretical Implications Now we have reached the conclusion that there are at least three ways of identification of null arguments in Japanese: (i) variables bound by null topics, (ii) bound pros, and (iii) null elements supplied with their content by the NP Ellipsis strategy. I would like to consider in this section where this conclusion leads us in any theoretical contexts. A significant point on this conclusion that should be pointed out in the first place is that no facts that have been considered so far point to the need of a referential pro, a pro that has the ability of referring; on the contrary, null arguments behave like dependent elements whose values need to be supplied from the outside. Given this characterization, it is natural to reason that null arguments in Japanese are not identified in the same way as those null subjects of such standard pro-drop languages as Italian and Spanish, which are typically analyzed as full-fledged pros. This, then, leads us to the question how null arguments in Japanese should be characterized within the overall theory of prodrop phenomena, or more concretely, what kind of parametrization is involved in characterizing null arguments in Japanese. I would like to suggest that the categorical size of noun phrases is relevant for the parametrization in question. In so doing, let us first outline relevant claims made by Holmberg (2005) with respect to third person definite subject pronouns in Finnish, which can be null only when they are bound by arguments in the next higher clauses, hence behaving like bound pros. Holmberg follows Déchaine and Wiltschko (2002) in classification of pronouns in terms of their categorial size, according to which they are either DPs, φPs, or NPs, but he exploits different criterions in the classification in question. He claims that the third person null pronoun in Finnish has the categorial status of φP in that it has φ-features that value those of T, but “lacks the substructure required for a definite, referential category.” (p. 552) Since they lack D, which is responsible for the ability of referring, they behave like pronouns that need to be bound to obtain their semantic values. Holmberg also claims that “the bound null pronoun and the generic null pronoun are the same category,” (ibid.) and that generic interpretation emerges when a null pronoun is bound by a null generic operator. Furthermore, Holmberg considers how to characterize null subjects in standard pro-drop languages such as Italian and Spanish (what he calls consistent null subject languages), which behave just like full-fledged overt pronouns in their referential properties. He proposes that “consistent null subject languages such as Spanish have a D-feature in I [=T], which is lacking in Finnish,” (p. 555) and that “presence of a D-feature in I means that a null φP

158　Jun Abe

that enters an Agree relation with I can be interpreted as definite, referring to an individual or a group.” (ibid.)12 Given this characterization of null pronouns, how do Japanese null arguments fit into the whole picture? Note first that they are similar to third person null subjects in Finnish in that they can be identified as bound pros. Furthermore, the sentences in (52) and (53) demonstrate that Japanese null arguments can also be interpreted as generic, another similarity to third person null subjects in Finnish. On the other hand, Japanese null arguments differ from those in Finnish in that they have access to the strategies of null topics and ellipsis. It may be natural to attribute this difference to the difference concerning the presence of φ-features; that is, since Japanese lacks φ-features, it can exploit different ways of identifying null arguments. Note, however, that the crucial finding of this article relevant for the present discussion is the fact that Japanese null arguments can exploit the same strategy of identification as those in Finnish, namely being identified as bound pros. Hence, it will be incorrect to simply draw a line between those languages which exploit φ-features in their grammatical systems and those which lack them and to claim that they involve totally different systems of identifying null arguments. I suggest that the lack of φ-features in a given language correlate with that of φ-projection in the nominal system of this language and hence that those null arguments which are observed in such a language have the categorial status of NP. Let us then hypothesize the following: (54) The ellipsis strategy is available only for an element that has no phonological, semantic, or formal features except a categorial feature. Suppose that null arguments in Japanese consist solely of nominal features. Then, they have access to the ellipsis strategy according to (54), and we have seen plenty of cases that demonstrate the need of such a strategy for null arguments in Japanese. Finnish null subject pronouns, on the other hand, do not seem to exploit such a strategy. The reason can be attributed to the fact that null subject pronouns in this language have the categorial status of φP and as such they enter into an Agree relation with T, which indicates that φ-features of these null arguments can act as more than simply categorial features, thereby 12

See Holmberg (2005) for detailed discussions on why the presence of a D-feature in I is responsible for the parametrization between consistent null subject languages and Finnish rather than the difference in categorial size between DP and φP. Note that Holmberg does not intend that there is any null argument that inherently has the categorial size of DP; in fact, he characterizes 1st and 2nd person null subjects in Finnish as DPs, since they can refer on their own.

Identification of Null Arguments in Japanese　159

prohibiting the null arguments from using the ellipsis strategy. That Japanese null arguments have only nominal features will be further supported by the above claim, based upon Hoji’s (1998) observation, that any portion of N’ projection can serve as an antecedent of a null argument in Japanese. Given Fukui and Spea’s (1986) theory of phrase structure, according to which lexical categories such as N, A and V have recursive projections of X’ with no closure of XP, it will not be unnatural to suppose that a null argument that has only a nominal feature accepts any N’ portion of a whole NP, including its head N, as its antecedent. So far, I have suggested that the absence of φ-features correlate with the availability of the ellipsis strategy. On the other hand, the fact that null arguments in both Finnish and Japanese can be identified as bound pros suggests that such identification is always available for any type of null arguments. To put those cases which involve identification by way of null topics to the side for the moment, we may depict the whole picture of identification of null arguments in somewhat figurative terms as follows: the most basic function of null arguments is to act as bound variables, which is genuinely manifested in null subject pronouns in Finnish and, plausibly, PRO in general.13 When a null argument obtains a D-feature from T, it behaves like a normal pronoun thanks to this feature providing it with the ability of referring. When it lacks φ-features, it has access to the ellipsis strategy, according to (54), while maintaining the ability of acting as a bound variable. Hence, identification of null arguments involves the following two parameters: (55) a. the availability of φ-features b. the availability of a D-feature in T If the above reasoning is on the right track, then the parameter (55a) works in accordance with Fukui’s (1986) theory of parametrization, according to which lack of φ-features correlate with that of functional categories. Finally, we are left with those cases which involve identification of null arguments by way of null topics in Japanese. I speculate that the availability of this identification is not directly related to the parametrization of null arguments per se but rather it is a special case of functioning as a bound variable; in this case, a variable bound by an operator such as a topic, minimally different from a variable bound by its antecedent in A-position. Hence, under this conception, the parametrization in question operates rather on the availability of null topics, as originally claimed by Huang (1982, 1984) in terms of Li and Thompson’s 13

See Abe (2007) for relevant discussions.

160　Jun Abe

(1976) topic vs. subject prominancy.

5. Conclusions and Some Relevant Remarks In this article, I first provided evidence for the NP Ellipsis strategy, advocated by Oku (1998) and Kim (1999), as a way of identification of Japanese null arguments. Then, I demonstrated that there need to be at least two further ways of identification: (i) variables bound by null topics and (ii) pros bound by NPs in A-position. From these observations, I drew the conclusion that Japanese null arguments differ from those of the standard pro-drop languages in that they lack the ability of referring. I suggested, based upon Holmberg’s (2005) analysis of null subjects in Finnish, that Japanese null arguments are peculiar in that they have the categorial status of NP, lacking functional categories, and that this will be the source of the inability of referring as well as the availability of the ellipsis strategy. Finally, I would like to make a brief mention about how to license null arguments. Here I am following Rizzi (1986) in distinguishing licensing from identification concerning null arguments. In Abe (2006), I hypothesized the following, based upon the ellipsis phenomena in Japanese: (56) Functional vs. Lexical Licensing Parameter If a language L exhibits agreement, then a null category is constrained by Lobeck’s (1995) licensing condition. If L lacks agreement, then a null category is licensed by selection of a lexical head. Lobeck (1995) follows Rizzi (1986) in assuming that genuine pro is licensed by head government14 and extends this to the licensing of such ellipsis phenomena as VP-Deletion, Sluicing and N’-Deletion. What I proposed with (56) is that in such languages as Chinese, Japanese and Korean which lack agreement, a different licensing condition is at work, according to which null arguments are licensed by way of lexical selection or predicate-argument relation. If this is the case, then it follows that the presence or lack of agreement affects both licensing and identification of null arguments in a given language. This is then another indication of how deeply agreement is involved in parametrization.

Identification of Null Arguments in Japanese　161

References Abe, Jun. 2006. Licensing conditions in ellipsis. Ms. Tohoku Gakuin University. Abe, Jun. 2007. Locality of empty pronouns in Japanese. Ms. Tohoku Gakuin University. Chao, Wynn 1987. On ellipsis. Doctoral dissertation. University of Massachusetts, Amherst. Chomsky, Noam. 1994. Bare phrase structure. MIT occasional papers in linguistics 5, MITWPL. Cambridge, Massachusetts. Déchaine, Rose-Marie and Martina Wiltschko. 2002. Decomposing pronouns. Linguistic Inquiry 33: 409-442. Evans, Gareth. 1980. Pronouns. Linguistic Inquiry 11: 337-362. Fox, Danny. 2000. Economy and semantic interpretation. Cambridge, Massachusetts: MIT Press. Fukui, Naoki. 1986. A Theory of category projection and its applications. Doctoral Dissertation. MIT. Fukui, Naoki and Margaret Speas. 1986. Specifiers and projection. In MIT working papers in linguistics 8, eds. N. Fukui, T.R. Rapoport and E. Sagey, 128-172. Hankamer, Jorge and Ivan Sag. 1976. Deep and surface anaphora. Linguistic Inquiry 7: 391-426. Hasegawa, Nobuko. 1984/5. On the so-called “zero pronouns” in Japanese. The Linguistic Review 4: 289-341. Hoji, Hajime. 1985. Logical form constraints and configurational structures in Japanese. Doctoral dissertation. University of Washington. Hoji, Hajime. 1998. Null object and sloppy identity in Japanese. Linguistic Inquiry 29: 127-152. Holmberg, Anders. 2005. Is there a little pro? Evidence from Finnish. Linguistic Inquiry 36: 533-564. Huang, C.-T. James. 1982. Logical relations in Chinese and the theory of grammar. Doctoral dissertation. MIT. Huang, C.-T. James. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry 15: 531-574. Huang, C.-T. James. 1991. Remarks on the status of the null object. In Principles and parameters in comparative grammar, ed. Robert Freidin, 56-76. Cambridge, Massachusetts: MIT Press. Kim, Soowon. 1999. Sloppy/strict identity, empty objects, and NP ellipsis. Journal of East Asian Linguistics 8: 255-284.

14

This licensing condition in terms of head government is responsible for the non-existence of pro in such languages as English and French, according to Rizzi (1986). Though such a formulation is clearly obsolete in the current framework of the Minimalist Program, I do not have anything interesting to say on this point.

Kuroda, S.-Y. 1965 Generative grammatical studies in the Japanese language. Doctoral dissertation. MIT. Kuroda, S.-Y. 1971. Remarks on the notion of subject with reference to words like

162　Jun Abe also, even or only. in The annual bulletin of the research institute of logopedics and phoniatrics 4, 127-152. University of Tokyo.

　163

A Note on Ellipsis of Quantificational Objects in Japanese

Li, Charles and Sandra Thompson. 1976. Subject and topic: A new typology of language. In Subject and topic, ed. Charles Li, 457-489. New York: Academic Press. Lobeck, Anne. 1995. Ellipsis: Functional heads, licensing, and identification. Oxford: Oxford University Press. Oku, Satoshi. 1998. A theory of selection and reconstruction in the minimalist perspective. Doctoral dissertation. University of Connecticut. Otani, Kazuyo and John Whitman. 1991. V-raising and VP-ellipsis. Linguistic Inquiry 22: 345-358. Reinhart, Tanya. 1998. Wh-in situ in the framework of the Minimalist Program. Natural Language Semantics 6: 29-56. Rizzi, Luigi. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501-557. Sag, Ivan. 1976. Deletion and logical form. Doctoral dissertation. MIT. Takahashi, Daiko. 2006. Apparent parasitic gaps and null arguments in Japanese. Journal of East Asian Linguistics 15: 1-35. Takahashi, Daiko. 2008. Quantificational null objects and argument ellipsis. Linguistic Inquiry 39: 307-326. Webber, Bonnie. 1978. A formal approach to discourse anaphora. Doctoral dissertation. Harvard University. Williams, Edwin. 1977. Discourse and logical form. Linguistic Inquiry 8: 103-139.

Daiko Takahashi 1. Introduction* Japanese has been known to be a language where arguments are allowed not to manifest themselves overtly. It has been standard since Kuroda (1965) to analyze them as phonetically unrealized pronouns. More recently, however, an alternative has been proposed that tries to analyze null arguments as involving ellipsis. The ellipsis analysis of null arguments has two varieties: One is due to Huang (1991) and Otani and Whitman (1991), who argue that the ellipsis operation in question is VP-ellipsis; the other is put forth by Kim (1999) and Oku (1998), who argue against the VP-ellipsis analysis (henceforth, the VPE analysis), instead proposing what we may call the argument ellipsis analysis (the AE analysis).1 The purpose of this article is to provide an argument that the AE analysis should be preferred to the VPE analysis on the basis of the null object construction where null objects are controlled by quantificational antecedents.

2. Null Objects: VP-Ellipsis or Argument Ellipsis? Let us start by considering the following examples: (1) a. Harii-ga zibun-no titioya-o hihansita. Harry-Nom self-Gen father-Acc criticized ‘lit. Harry criticized self’s father.’ b. Ron-mo e hihansita. Ron-also criticized ‘lit. Ron criticized, too.’ Sentence (1a) serves as the antecedent for (1b), where the object is empty. While (1a) means only that Harry criticized his own father (namely, Harry’s father), (1b) is ambiguous with respect to the interpretation of the null object: it means either that Ron criticized Harry’s father or that Ron criticized his own (namely, Ron’s) father (let us call these readings the strict reading and the sloppy reading, respectively). The presence of the strict interpretation is expected under the view *

This article is partially based on the material I presented at the International Symposium on Language, Mind, and Brain, held at Akita University in October, 2006. For their helpful comments and questions I would like to thank the participants, especially Jun Abe, Hiroto Hoshi, Ruth Kempson. Jieun Joe Kiaer, and Masatoshi Koizumi. 1 The term argument ellipsis is coined by Saito (2004).